MCP vs Google's Agent2Agent - one to watch

Like many, I've been playing around with Anthropic's MCP (Model-Context-Protocol) with my Claude Desktop App and I have occasionally been pretty impressed with it. I've extended Claude Desktop with a whole load of different MCP capabilities both for research purposes and to try and actively use the capability properly.

Aside from having to constantly grant various different security approvals I found the underlying reality to be a little bit limited, so far. A little bit Fisher Price, as the phrase goes. It doesn't necessarily feel anywhere near enterprise-grade yet. And, look, I know that we're in the ultra-early stages of extending the likes of Claude or ChatGPT to do more.

Having heard of various different rumblings from Google recently, it's great to see their announcement of Agent2Agent or "A2A". It's an open protocol designed to enable AI agents to collaborate across different systems, applications, and vendors.

There's a who's-who list of supporting firms from Accenture to Box to KPMG to Oracle, Salesforce, Datadog, Salesforce and Wipro – just to throw out a few names.

The design principles are as you might expect - secure by default, uses existing standards (e.g. HTTP/JSON) and interestingly, will offer 'long-running' task support, rather than tasks that are expected to be completed in (mili)seconds.

I think this is definitely one to watch closely. Have a glance over the announcement post and keep an eye on it.

Plugging the glue between the LLM, the end user and all the various different services and capabilities out there is going to be really, really important.

I won't be surprised if the adoption of this kind of capability happens at light speed, especially with (for example) 500M people using the likes of ChatGPT every week.

If you're anywhere near playing around with Agentic or chatbot capabilities, I think you or someone in your company should be all over this.